CTN Seminar: Nachi Stern: Learning in physical machines
Interacting many-body physical systems ranging from electrically responsive neural networks to folding proteins can learn to perform specific tasks. This learning, both in nature and in engineered systems, can occur through evolutionary selection or through dynamical rules that drive active learning from experience. We show that learning leaves architectural signatures in the physical system. Compared to a generic organization of the system components, (a) the effective physical dimension decreases, (b) the dynamical range of responses (or "effective conductance") increases, and (c) the inherent coordinate system of dynamics realigns to the task. Overall, these effects suggest a method for discovering the task that a novel physical network may have been trained for.